Institute of Information Science, Academia Sinica

Abstract: We propose a method to generate a highly accurate 3-D face model from a set of wide-baseline images in a weakly calibrated setup. Our approach is purely data driven, and produces faithful 3-D models without any pre-defined models, unlike other statistical model-based approaches. Our results do not rely upon a critical initialization step nor parameters for optimization steps. We process 5 images (including profile views), infer the accurate poses of cameras in all views, and then infer a dense 3-D face model. The quality of 3-D face models depends on the accuracy of estimated head-camera motion. First, we propose to use an iterative bundle adjustment approach to remove outliers in corresponding points. Contours in the profile views are matched to provide reliable correspondences that link two opposite side of views together. For dense reconstruction, we propose to use a face-specific cylindrical representation which allows us to solve a global optimization problem for N-view dense aggregation. Profile contours are used once again to provide constraints in the optimization step. Experimental results using synthetic and real images show that our method provides accurate and stable reconstruction results on wide-baseline images. We compare our method with state of the art methods, and show that it provides significantly better results in terms of both accuracy and efficiency. In addition, we also developed an algorithm for efficient and accurate 3D reconstruction of urban scene. We present a novel approach to perform 3D reconstruction in urban scenes from aerial imagery. State-of-the art 3D urban scene reconstruction methods use Manhattan world assumption to regularize the structure (scene is piecewise planar, and planes are axis aligned), which is only valid for ground-based images. Our approach, on the other hand, makes a more general assumption that the planes are either horizontal or vertical. Along with edge information that is prevalent in urban scene imagery, we formulate the dense reconstruction problem as a 2-pass dynamic programming problem, which can be solved efficiently. Moreover, our algorithm is fully parallelizable which performs the reconstruction of 1M points (with 160 discrete height levels) in less than 1 minute on a GPU. Results preserve high level of detail and show high visual quality. Bio: Yuping Lin received the B.B.A degree in information management from National Taiwan University, Taipei, and the M.S. degree in computer science from University of Southern California, Los Angeles, in 2003 and 2006, respectively. He is currently pursuing the Ph.D. degree in University of Southern California, Los Angeles. He was in Institute of Information Science, Academia Sinica, Taiwan, in year 2003 and 2005 spring, and in Siemens Medical Solutions, Philadelphia, in year 2008 summer. His recent research interests include image registration and in 3D reconstruction.